Add thousand separators to meshing utility output

Functionality to add/problem to solve

Adding thousand separators to the output so it's easier for human to read long integers.

Target audience

All meshing utilities (and probably some other applications)

Proposal

Add some code to messageStream when printing integers

What does success look like, and how can we measure that?

For example the current output of blockMesh is like

Mesh Information
----------------
  boundingBox: (0 -0.0005 0) (0.012 0.0005 5e-05)
  nPoints: 4911759
  nCells: 4729600
  nFaces: 14369360
  nInternalFaces: 14008240
----------------

Adding thousand separators to integers will make it much easier to read.

Mesh Information
----------------
  boundingBox: (0 -0.0005 0) (0.012 0.0005 5e-05)
  nPoints: 4,911,759
  nCells: 4,729,600
  nFaces: 14,369,360
  nInternalFaces: 14,008,240
----------------

Links / references

You don't need to write any code to parse numbers. C++ has built-in functions to do that. https://cplusplus.com/reference/locale/numpunct/

changed the description

added community feature request labels

I don't quite follow how this could be added to messageStream only. A messageStream like Info has an implicit conversion to OSstream which wraps either a std::cout or a /dev/null. Would need to add an extra inheritance layer to fit in the formatting (how?) or have a duplicate of Sout with the special imbue locale stuff.

I personally don't really like the comma as a thousands separator, having fought too many times with locales like German - eg, €1.000,00 vs $1,000.00. Did always like the Perl separator _` which would actually be possible to build back into our own parser, but interesting enough C++14 uses a single quote - wouldn't want to start parsing that!

If we want this type of thing (I'm not sure I do), could try with an IOmanip operator. The output region would probably looks like this:

Info<< thousands_comma << "Have " << nCells << " cells" << nl
    << thousands_plain;

but this is getting ugly enough that we probably need to have namespace prefix.

For testing purposes, I think this type of definition would be doable within a small test code (in a single file). Would you mind having a go at that?

Cheers,

/mark

Here without proper wrapping: Test-locale.C

Thanks. I didn't consider the complex situation in Europe. And in east Asia, numbers are grouped every ten thousand (orally). If it's too complicated to implement based on locale, then it's better keep the current format.

OK, I'll close it. For programming, any locale other than "C" gets really annoying since things like std::to_string() actually respects locale (eg, PI = 3,14159 for DE-de locale, but 3.14159 for C and US-en locale) but std::ostringstream does not. For Linux itself, it's a bit of a mixed bag. It gets even more fun with things like UK-English settings on German keyboard, but sometimes the lowest level doesn't know that the keyboard is German so avoid any special characters on passwords if you don't know where they appear on both keyboard mappings.

closed